Personas are, for my money, the most misused and abused tool in the CX toolkit. I have built hundreds of them over the course of my career, sat in workshops where they were treated as gospel, and watched them gather dust on walls within weeks of the project ending. So, when I was asked whether customer personas are still relevant in a world where agentic AI is reshaping how customers behave and buy, I had a lot of thoughts.
The short answer is they were always wobblier than we admitted. Machine customers and agentic AI have simply made that visible.
The original problem with personas
The classic critique has been circulating in CX and UX circles for years. Two men, both born in 1948, both wealthy, both fond of winter sports, married twice, fathers, dog owners, raised in England. These identical demographic descriptors describe both King Charles and Ozzy Osbourne. Two people who are, let's be fair, completely different behavioral profiles and should be treated very differently as customers, regardless of what you are selling.
Demographic and psychographic slices, when applied bluntly, reinforce stereotypes rather than illuminate behaviour.
When done well, personas help teams resist the temptation to design for themselves, build shared understanding, and keep real human beings in the room during decisions that affect them. That is genuine value. The problem is that "when done well" has always been the exception rather than the rule.
Where machine customers break the tool entirely
When I started exploring machine customers and agentic commerce, personas seemed like the right tool to reach for. Rather than treating the arrival of AI procurement agents as a purely technical problem to be solved with APIs and data lakes, I thought personas could help us design for these unfamiliar customers using a familiar framework.
I took that idea to customer research expert Indi Young for a stress test. Her response cut straight to the problem:
"I would not use the word 'persona', okay, call them 'types'. They're just types of agents, right? A persona is supposed to represent preferences; they represent ways of thinking. And that is not true of agents. It is an entity like Siri, your Google assistant or whatever that executes commands and does tasks for you, and what you're describing isn't its preferences around doing those tasks, but the tasks themselves."
Ouch. My poor idea baby. She was right, and I knew it. The word "persona" carries the assumption of consciousness, preference and human-like psychology. Machine customers have none of those things. I talk often about “Tyler”, a delegated AI agent, who could be shopping for ergonomic chairs with sustainability certifications and machine-readable warranties, is thinking, feeling, seeing, and hearing nothing. It is processing parameters and executing logic. An empathy map produces a blank page. A persona produces a fiction.
What works instead are ‘types'. That is to say ‘functional categories based on the tasks being performed, the scope of authority involved, and the interaction patterns required’. A delegated agent like Tyler and an autonomous enterprise procurement system are different types entirely, with different CX needs. The distinction is about task complexity and decision-making parameters, not personality.
The pressure from the human side
The provocation on personas extended beyond machine customers to something equally worth examining. Agentic AI within organizations is also creating dynamic journeys based on human behaviour, reshaping how even human customers move through consideration and purchase.
A persona built on a static demographic or psychographic profile assumes a relatively predictable customer who follows a roughly mappable path.
When an AI agent is acting on that human's behalf, screening vendors, filtering options, and presenting recommendations before the human ever gets involved, the persona you built is describing someone who may never directly touch your experience at all.
The human behind Tyler matters enormously. Their preferences, values and priorities are encoded in Tyler's parameters. But Tyler itself is not a persona candidate. And the human whose decisions are being filtered through Tyler is increasingly invisible in the early stages of the purchase journey, which is exactly where traditional personas were meant to do their heaviest lifting.
Thinking styles and the limits of reframing
I also explored whether Indi Young's concept of thinking styles – a framework developed as a more behaviorally grounded alternative to traditional personas – might apply to machine customers. Indi describes thinking styles as being "free of demographics", focused instead on "what's actually going on in people's minds" as they pursue a goal.
The attempt also did not survive contact with the concept. Machine customers have no minds, no goals in the human sense, and none of the contextual variability that thinking styles are designed to capture.
What became clear through that conversation was that the useful thing is understanding the task sets machine customers execute, and designing service protocols around task complexity.
That is a genuinely different cognitive operation from persona work, and trying to force one framework into the shape of the other produces neither.
What the persona was actually doing for us
I don’t think we need to ask whether to discard personas, but we do need to understand clearly what cognitive job they were doing. They were keeping real humans in the room.
They were pushing teams away from designing for themselves. They were making customer diversity visible in organizations that would otherwise default to a single imagined user.
Those jobs still need doing. For human customers, better behavioral research and task-based frameworks will serve teams more accurately than demographic proxies, particularly when an AI agent is mediating the experience before the human appears. For machine customers, the tool that replaces the persona is the type: a functional profile built around tasks, authority levels, decision logic and integration requirements.
The persona retains a clearly defined, still-valuable role in human-centered design. It simply was never built to do what we are now asking of it. And the sooner we stop asking, the sooner we can design experiences for AI agents that actually work.
Quick links
- The human-machine translator: CX's most important new role
- Agentic Agility: The new ‘normal’ for CX in 2026
- There’s no point in great CX if it doesn’t create revenue